Clinical Trials
AI-enabled analysis of pathology slides may help assess risk of pancreatic cancer recurrence
BroadfootMayo Clinic

Mayo Clinic investigators have developed an artificial intelligence (AI)-enabled spatial analysis platform capable of evaluating routine hematoxylin and eosin (H&E) pathology slides to predict disease recurrence in patients with pancreatic ductal adenocarcinoma. Published in Clinical Cancer Research, the study evaluated surgical tissue from 203 patients who underwent neoadjuvant therapy followed by surgical resection but demonstrated only a limited pathologic response.
Conventional postoperative pathologic evaluation relies primarily on quantifying the burden of residual viable tumor. However, the study demonstrated that residual tumor volume alone fails to reliably discriminate between patients at high versus low risk of recurrence. By applying principles adapted from landscape ecology and machine learning to digitized standard H&E sections, researchers quantified the architectural configuration, boundary patterns, fragmentation, and intermixing between residual malignant epithelial glands and the surrounding desmoplastic stroma. Patients whose residual disease exhibited highly fragmented and intermixed cancer-stroma geography experienced significantly earlier disease recurrence.
Multivariate analysis demonstrated that two distinct AI-derived spatial signatures independently predicted disease-free survival after controlling for established clinical and pathologic covariates, including disease stage and lymph node status. Patients stratified into high-risk categories exhibited a 71% higher adjusted recurrence risk in one model and more than twice the adjusted risk in another. Biologically, high-risk spatial configurations correlated with altered tumor microenvironments characterized by immune exclusion, wherein immune cell populations localized at the tumor margins rather than infiltrating the core malignant tissue. Because this spatial approach leverages existing routine H&E histology, it offers a potential pathway to enhance prognostic precision and risk-stratified surveillance without requiring additional tissue testing, though prospective validation remains necessary before clinical deployment.
Key Takeaways
- In multivariate models adjusting for stage, nodal status, and established pathologic factors, high-risk spatial tumor-stroma patterns independently predicted a 71% to greater than twofold increase in adjusted recurrence risk following resection.
- Intratumoral immune cell infiltration was significantly reduced in tumors exhibiting high-risk spatial architecture, with immune cells predominantly restricted to the tumor periphery rather than penetrating the malignant tissue.
- AI-driven spatial metrics of tissue fragmentation and cancer-stroma intermixing successfully stratified disease-free survival in patients with limited neoadjuvant response, whereas residual tumor volume alone failed to distinguish risk.
Read the original sourceMayo Clinic
